Copyright updates:
[exim.git] / src / src / spam.c
1 /*************************************************
2 * Exim - an Internet mail transport agent *
3 *************************************************/
4
5 /* Copyright (c) Tom Kistner <tom@duncanthrax.net> 2003 - 2015
6 * License: GPL
7 * Copyright (c) The Exim Maintainers 2016 - 2020
8 */
9
10 /* Code for calling spamassassin's spamd. Called from acl.c. */
11
12 #include "exim.h"
13 #ifdef WITH_CONTENT_SCAN
14 #include "spam.h"
15
16 uschar spam_score_buffer[16];
17 uschar spam_score_int_buffer[16];
18 uschar spam_bar_buffer[128];
19 uschar spam_action_buffer[32];
20 uschar spam_report_buffer[32600];
21 uschar prev_user_name[128] = "";
22 int spam_ok = 0;
23 int spam_rc = 0;
24 uschar *prev_spamd_address_work = NULL;
25
26 static const uschar * loglabel = US"spam acl condition:";
27
28
29 static int
30 spamd_param_init(spamd_address_container *spamd)
31 {
32 /* default spamd server weight, time and priority value */
33 spamd->is_rspamd = FALSE;
34 spamd->is_failed = FALSE;
35 spamd->weight = SPAMD_WEIGHT;
36 spamd->timeout = SPAMD_TIMEOUT;
37 spamd->retry = 0;
38 spamd->priority = 1;
39 return 0;
40 }
41
42
43 static int
44 spamd_param(const uschar * param, spamd_address_container * spamd)
45 {
46 static int timesinceday = -1;
47 const uschar * s;
48 const uschar * name;
49
50 /*XXX more clever parsing could discard embedded spaces? */
51
52 if (sscanf(CCS param, "pri=%u", &spamd->priority))
53 return 0; /* OK */
54
55 if (sscanf(CCS param, "weight=%u", &spamd->weight))
56 {
57 if (spamd->weight == 0) /* this server disabled: skip it */
58 return 1;
59 return 0; /* OK */
60 }
61
62 if (Ustrncmp(param, "time=", 5) == 0)
63 {
64 unsigned int start_h = 0, start_m = 0, start_s = 0;
65 unsigned int end_h = 24, end_m = 0, end_s = 0;
66 unsigned int time_start, time_end;
67 const uschar * end_string;
68
69 name = US"time";
70 s = param+5;
71 if ((end_string = Ustrchr(s, '-')))
72 {
73 end_string++;
74 if ( sscanf(CS end_string, "%u.%u.%u", &end_h, &end_m, &end_s) == 0
75 || sscanf(CS s, "%u.%u.%u", &start_h, &start_m, &start_s) == 0
76 )
77 goto badval;
78 }
79 else
80 goto badval;
81
82 if (timesinceday < 0)
83 {
84 time_t now = time(NULL);
85 struct tm *tmp = localtime(&now);
86 timesinceday = tmp->tm_hour*3600 + tmp->tm_min*60 + tmp->tm_sec;
87 }
88
89 time_start = start_h*3600 + start_m*60 + start_s;
90 time_end = end_h*3600 + end_m*60 + end_s;
91
92 if (timesinceday < time_start || timesinceday >= time_end)
93 return 1; /* skip spamd server */
94
95 return 0; /* OK */
96 }
97
98 if (Ustrcmp(param, "variant=rspamd") == 0)
99 {
100 spamd->is_rspamd = TRUE;
101 return 0;
102 }
103
104 if (Ustrncmp(param, "tmo=", 4) == 0)
105 {
106 int sec = readconf_readtime((s = param+4), '\0', FALSE);
107 name = US"timeout";
108 if (sec < 0)
109 goto badval;
110 spamd->timeout = sec;
111 return 0;
112 }
113
114 if (Ustrncmp(param, "retry=", 6) == 0)
115 {
116 int sec = readconf_readtime((s = param+6), '\0', FALSE);
117 name = US"retry";
118 if (sec < 0)
119 goto badval;
120 spamd->retry = sec;
121 return 0;
122 }
123
124 log_write(0, LOG_MAIN, "%s warning - invalid spamd parameter: '%s'",
125 loglabel, param);
126 return -1; /* syntax error */
127
128 badval:
129 log_write(0, LOG_MAIN,
130 "%s warning - invalid spamd %s value: '%s'", loglabel, name, s);
131 return -1; /* syntax error */
132 }
133
134
135 static int
136 spamd_get_server(spamd_address_container ** spamds, int num_servers)
137 {
138 unsigned int i;
139 spamd_address_container * sd;
140 long weights;
141 unsigned pri;
142 static BOOL srandomed = FALSE;
143
144 /* speedup, if we have only 1 server */
145 if (num_servers == 1)
146 return (spamds[0]->is_failed ? -1 : 0);
147
148 /* init ranmod */
149 if (!srandomed)
150 {
151 struct timeval tv;
152 gettimeofday(&tv, NULL);
153 srandom((unsigned int)(tv.tv_usec/1000));
154 srandomed = TRUE;
155 }
156
157 /* scan for highest pri */
158 for (pri = 0, i = 0; i < num_servers; i++)
159 {
160 sd = spamds[i];
161 if (!sd->is_failed && sd->priority > pri) pri = sd->priority;
162 }
163
164 /* get sum of weights */
165 for (weights = 0, i = 0; i < num_servers; i++)
166 {
167 sd = spamds[i];
168 if (!sd->is_failed && sd->priority == pri) weights += sd->weight;
169 }
170 if (weights == 0) /* all servers failed */
171 return -1;
172
173 for (long rnd = random() % weights, i = 0; i < num_servers; i++)
174 {
175 sd = spamds[i];
176 if (!sd->is_failed && sd->priority == pri)
177 if ((rnd -= sd->weight) <= 0)
178 return i;
179 }
180
181 log_write(0, LOG_MAIN|LOG_PANIC,
182 "%s unknown error (memory/cpu corruption?)", loglabel);
183 return -1;
184 }
185
186
187 int
188 spam(const uschar **listptr)
189 {
190 int sep = 0;
191 const uschar *list = *listptr;
192 uschar *user_name;
193 uschar user_name_buffer[128];
194 unsigned long mbox_size;
195 FILE *mbox_file;
196 client_conn_ctx spamd_cctx = {.sock = -1};
197 uschar spamd_buffer[32600];
198 int i, j, offset, result;
199 uschar spamd_version[8];
200 uschar spamd_short_result[8];
201 uschar spamd_score_char;
202 double spamd_threshold, spamd_score, spamd_reject_score;
203 int spamd_report_offset;
204 uschar *p,*q;
205 int override = 0;
206 time_t start;
207 size_t read, wrote;
208 #ifndef NO_POLL_H
209 struct pollfd pollfd;
210 #else /* Patch posted by Erik ? for OS X */
211 struct timeval select_tv; /* and applied by PH */
212 fd_set select_fd;
213 #endif
214 uschar *spamd_address_work;
215 spamd_address_container * sd;
216
217 /* stop compiler warning */
218 result = 0;
219
220 /* find the username from the option list */
221 if ((user_name = string_nextinlist(&list, &sep,
222 user_name_buffer,
223 sizeof(user_name_buffer))) == NULL)
224 {
225 /* no username given, this means no scanning should be done */
226 return FAIL;
227 }
228
229 /* if username is "0" or "false", do not scan */
230 if ( (Ustrcmp(user_name,"0") == 0) ||
231 (strcmpic(user_name,US"false") == 0) )
232 return FAIL;
233
234 /* if there is an additional option, check if it is "true" */
235 if (strcmpic(list,US"true") == 0)
236 /* in that case, always return true later */
237 override = 1;
238
239 /* expand spamd_address if needed */
240 if (*spamd_address == '$')
241 {
242 spamd_address_work = expand_string(spamd_address);
243 if (spamd_address_work == NULL)
244 {
245 log_write(0, LOG_MAIN|LOG_PANIC,
246 "%s spamd_address starts with $, but expansion failed: %s",
247 loglabel, expand_string_message);
248 return DEFER;
249 }
250 }
251 else
252 spamd_address_work = spamd_address;
253
254 DEBUG(D_acl) debug_printf_indent("spamd: addrlist '%s'\n", spamd_address_work);
255
256 /* check if previous spamd_address was expanded and has changed. dump cached results if so */
257 if ( spam_ok
258 && prev_spamd_address_work != NULL
259 && Ustrcmp(prev_spamd_address_work, spamd_address_work) != 0
260 )
261 spam_ok = 0;
262
263 /* if we scanned for this username last time, just return */
264 if (spam_ok && Ustrcmp(prev_user_name, user_name) == 0)
265 return override ? OK : spam_rc;
266
267 /* make sure the eml mbox file is spooled up */
268
269 if (!(mbox_file = spool_mbox(&mbox_size, NULL, NULL)))
270 { /* error while spooling */
271 log_write(0, LOG_MAIN|LOG_PANIC,
272 "%s error while creating mbox spool file", loglabel);
273 return DEFER;
274 }
275
276 start = time(NULL);
277
278 {
279 int num_servers = 0;
280 int current_server;
281 uschar * address;
282 const uschar * spamd_address_list_ptr = spamd_address_work;
283 spamd_address_container * spamd_address_vector[32];
284
285 /* Check how many spamd servers we have
286 and register their addresses */
287 sep = 0; /* default colon-sep */
288 while ((address = string_nextinlist(&spamd_address_list_ptr, &sep, NULL, 0)))
289 {
290 const uschar * sublist;
291 int sublist_sep = -(int)' '; /* default space-sep */
292 unsigned args;
293 uschar * s;
294
295 DEBUG(D_acl) debug_printf_indent("spamd: addr entry '%s'\n", address);
296 sd = store_get(sizeof(spamd_address_container), FALSE);
297
298 for (sublist = address, args = 0, spamd_param_init(sd);
299 (s = string_nextinlist(&sublist, &sublist_sep, NULL, 0));
300 args++
301 )
302 {
303 DEBUG(D_acl) debug_printf_indent("spamd: addr parm '%s'\n", s);
304 switch (args)
305 {
306 case 0: sd->hostspec = s;
307 if (*s == '/') args++; /* local; no port */
308 break;
309 case 1: sd->hostspec = string_sprintf("%s %s", sd->hostspec, s);
310 break;
311 default: spamd_param(s, sd);
312 break;
313 }
314 }
315 if (args < 2)
316 {
317 log_write(0, LOG_MAIN,
318 "%s warning - invalid spamd address: '%s'", loglabel, address);
319 continue;
320 }
321
322 spamd_address_vector[num_servers] = sd;
323 if (++num_servers > 31)
324 break;
325 }
326
327 /* check if we have at least one server */
328 if (!num_servers)
329 {
330 log_write(0, LOG_MAIN|LOG_PANIC,
331 "%s no useable spamd server addresses in spamd_address configuration option.",
332 loglabel);
333 goto defer;
334 }
335
336 current_server = spamd_get_server(spamd_address_vector, num_servers);
337 sd = spamd_address_vector[current_server];
338 for(;;)
339 {
340 uschar * errstr;
341
342 DEBUG(D_acl) debug_printf_indent("spamd: trying server %s\n", sd->hostspec);
343
344 for (;;)
345 {
346 /*XXX could potentially use TFO early-data here */
347 if ( (spamd_cctx.sock = ip_streamsocket(sd->hostspec, &errstr, 5, NULL)) >= 0
348 || sd->retry <= 0
349 )
350 break;
351 DEBUG(D_acl) debug_printf_indent("spamd: server %s: retry conn\n", sd->hostspec);
352 while (sd->retry > 0) sd->retry = sleep(sd->retry);
353 }
354 if (spamd_cctx.sock >= 0)
355 break;
356
357 log_write(0, LOG_MAIN, "%s spamd: %s", loglabel, errstr);
358 sd->is_failed = TRUE;
359
360 current_server = spamd_get_server(spamd_address_vector, num_servers);
361 if (current_server < 0)
362 {
363 log_write(0, LOG_MAIN|LOG_PANIC, "%s all spamd servers failed", loglabel);
364 goto defer;
365 }
366 sd = spamd_address_vector[current_server];
367 }
368 }
369
370 (void)fcntl(spamd_cctx.sock, F_SETFL, O_NONBLOCK);
371 /* now we are connected to spamd on spamd_cctx.sock */
372 if (sd->is_rspamd)
373 {
374 gstring * req_str;
375 const uschar * s;
376
377 req_str = string_append(NULL, 8,
378 "CHECK RSPAMC/1.3\r\nContent-length: ", string_sprintf("%lu\r\n", mbox_size),
379 "Queue-Id: ", message_id,
380 "\r\nFrom: <", sender_address,
381 ">\r\nRecipient-Number: ", string_sprintf("%d\r\n", recipients_count));
382
383 for (int i = 0; i < recipients_count; i++)
384 req_str = string_append(req_str, 3,
385 "Rcpt: <", recipients_list[i].address, ">\r\n");
386 if ((s = expand_string(US"$sender_helo_name")) && *s)
387 req_str = string_append(req_str, 3, "Helo: ", s, "\r\n");
388 if ((s = expand_string(US"$sender_host_name")) && *s)
389 req_str = string_append(req_str, 3, "Hostname: ", s, "\r\n");
390 if (sender_host_address)
391 req_str = string_append(req_str, 3, "IP: ", sender_host_address, "\r\n");
392 if ((s = expand_string(US"$authenticated_id")) && *s)
393 req_str = string_append(req_str, 3, "User: ", s, "\r\n");
394 req_str = string_catn(req_str, US"\r\n", 2);
395 wrote = send(spamd_cctx.sock, req_str->s, req_str->ptr, 0);
396 }
397 else
398 { /* spamassassin variant */
399 (void)string_format(spamd_buffer,
400 sizeof(spamd_buffer),
401 "REPORT SPAMC/1.2\r\nUser: %s\r\nContent-length: %ld\r\n\r\n",
402 user_name,
403 mbox_size);
404 /* send our request */
405 wrote = send(spamd_cctx.sock, spamd_buffer, Ustrlen(spamd_buffer), 0);
406 }
407
408 if (wrote == -1)
409 {
410 (void)close(spamd_cctx.sock);
411 log_write(0, LOG_MAIN|LOG_PANIC,
412 "%s spamd %s send failed: %s", loglabel, callout_address, strerror(errno));
413 goto defer;
414 }
415
416 /* now send the file */
417 /* spamd sometimes accepts connections but doesn't read data off
418 * the connection. We make the file descriptor non-blocking so
419 * that the write will only write sufficient data without blocking
420 * and we poll the descriptor to make sure that we can write without
421 * blocking. Short writes are gracefully handled and if the whole
422 * transaction takes too long it is aborted.
423 * Note: poll() is not supported in OSX 10.2 and is reported to be
424 * broken in more recent versions (up to 10.4).
425 */
426 #ifndef NO_POLL_H
427 pollfd.fd = spamd_cctx.sock;
428 pollfd.events = POLLOUT;
429 #endif
430 (void)fcntl(spamd_cctx.sock, F_SETFL, O_NONBLOCK);
431 do
432 {
433 read = fread(spamd_buffer,1,sizeof(spamd_buffer),mbox_file);
434 if (read > 0)
435 {
436 offset = 0;
437 again:
438 #ifndef NO_POLL_H
439 result = poll(&pollfd, 1, 1000);
440
441 /* Patch posted by Erik ? for OS X and applied by PH */
442 #else
443 select_tv.tv_sec = 1;
444 select_tv.tv_usec = 0;
445 FD_ZERO(&select_fd);
446 FD_SET(spamd_cctx.sock, &select_fd);
447 result = select(spamd_cctx.sock+1, NULL, &select_fd, NULL, &select_tv);
448 #endif
449 /* End Erik's patch */
450
451 if (result == -1 && errno == EINTR)
452 goto again;
453 else if (result < 1)
454 {
455 if (result == -1)
456 log_write(0, LOG_MAIN|LOG_PANIC,
457 "%s %s on spamd %s socket", loglabel, callout_address, strerror(errno));
458 else
459 {
460 if (time(NULL) - start < sd->timeout)
461 goto again;
462 log_write(0, LOG_MAIN|LOG_PANIC,
463 "%s timed out writing spamd %s, socket", loglabel, callout_address);
464 }
465 (void)close(spamd_cctx.sock);
466 goto defer;
467 }
468
469 wrote = send(spamd_cctx.sock,spamd_buffer + offset,read - offset,0);
470 if (wrote == -1)
471 {
472 log_write(0, LOG_MAIN|LOG_PANIC,
473 "%s %s on spamd %s socket", loglabel, callout_address, strerror(errno));
474 (void)close(spamd_cctx.sock);
475 goto defer;
476 }
477 if (offset + wrote != read)
478 {
479 offset += wrote;
480 goto again;
481 }
482 }
483 }
484 while (!feof(mbox_file) && !ferror(mbox_file));
485
486 if (ferror(mbox_file))
487 {
488 log_write(0, LOG_MAIN|LOG_PANIC,
489 "%s error reading spool file: %s", loglabel, strerror(errno));
490 (void)close(spamd_cctx.sock);
491 goto defer;
492 }
493
494 (void)fclose(mbox_file);
495
496 /* we're done sending, close socket for writing */
497 if (!sd->is_rspamd)
498 shutdown(spamd_cctx.sock,SHUT_WR);
499
500 /* read spamd response using what's left of the timeout. */
501 memset(spamd_buffer, 0, sizeof(spamd_buffer));
502 offset = 0;
503 while ((i = ip_recv(&spamd_cctx,
504 spamd_buffer + offset,
505 sizeof(spamd_buffer) - offset - 1,
506 sd->timeout + start)) > 0)
507 offset += i;
508 spamd_buffer[offset] = '\0'; /* guard byte */
509
510 /* error handling */
511 if (i <= 0 && errno != 0)
512 {
513 log_write(0, LOG_MAIN|LOG_PANIC,
514 "%s error reading from spamd %s, socket: %s", loglabel, callout_address, strerror(errno));
515 (void)close(spamd_cctx.sock);
516 return DEFER;
517 }
518
519 /* reading done */
520 (void)close(spamd_cctx.sock);
521
522 if (sd->is_rspamd)
523 { /* rspamd variant of reply */
524 int r;
525 if ( (r = sscanf(CS spamd_buffer,
526 "RSPAMD/%7s 0 EX_OK\r\nMetric: default; %7s %lf / %lf / %lf\r\n%n",
527 spamd_version, spamd_short_result, &spamd_score, &spamd_threshold,
528 &spamd_reject_score, &spamd_report_offset)) != 5
529 || spamd_report_offset >= offset /* verify within buffer */
530 )
531 {
532 log_write(0, LOG_MAIN|LOG_PANIC,
533 "%s cannot parse spamd %s, output: %d", loglabel, callout_address, r);
534 return DEFER;
535 }
536 /* now parse action */
537 p = &spamd_buffer[spamd_report_offset];
538
539 if (Ustrncmp(p, "Action: ", sizeof("Action: ") - 1) == 0)
540 {
541 p += sizeof("Action: ") - 1;
542 q = &spam_action_buffer[0];
543 while (*p && *p != '\r' && (q - spam_action_buffer) < sizeof(spam_action_buffer) - 1)
544 *q++ = *p++;
545 *q = '\0';
546 }
547 }
548 else
549 { /* spamassassin */
550 /* dig in the spamd output and put the report in a multiline header,
551 if requested */
552 if (sscanf(CS spamd_buffer,
553 "SPAMD/%7s 0 EX_OK\r\nContent-length: %*u\r\n\r\n%lf/%lf\r\n%n",
554 spamd_version,&spamd_score,&spamd_threshold,&spamd_report_offset) != 3)
555 {
556 /* try to fall back to pre-2.50 spamd output */
557 if (sscanf(CS spamd_buffer,
558 "SPAMD/%7s 0 EX_OK\r\nSpam: %*s ; %lf / %lf\r\n\r\n%n",
559 spamd_version,&spamd_score,&spamd_threshold,&spamd_report_offset) != 3)
560 {
561 log_write(0, LOG_MAIN|LOG_PANIC,
562 "%s cannot parse spamd %s output", loglabel, callout_address);
563 return DEFER;
564 }
565 }
566
567 Ustrcpy(spam_action_buffer,
568 spamd_score >= spamd_threshold ? US"reject" : US"no action");
569 }
570
571 /* Create report. Since this is a multiline string,
572 we must hack it into shape first */
573 p = &spamd_buffer[spamd_report_offset];
574 q = spam_report_buffer;
575 while (*p != '\0')
576 {
577 /* skip \r */
578 if (*p == '\r')
579 {
580 p++;
581 continue;
582 }
583 *q++ = *p;
584 if (*p++ == '\n')
585 {
586 /* add an extra space after the newline to ensure
587 that it is treated as a header continuation line */
588 *q++ = ' ';
589 }
590 }
591 /* NULL-terminate */
592 *q-- = '\0';
593 /* cut off trailing leftovers */
594 while (*q <= ' ')
595 *q-- = '\0';
596
597 spam_report = spam_report_buffer;
598 spam_action = spam_action_buffer;
599
600 /* create spam bar */
601 spamd_score_char = spamd_score > 0 ? '+' : '-';
602 j = abs((int)(spamd_score));
603 i = 0;
604 if (j != 0)
605 while ((i < j) && (i <= MAX_SPAM_BAR_CHARS))
606 spam_bar_buffer[i++] = spamd_score_char;
607 else
608 {
609 spam_bar_buffer[0] = '/';
610 i = 1;
611 }
612 spam_bar_buffer[i] = '\0';
613 spam_bar = spam_bar_buffer;
614
615 /* create "float" spam score */
616 (void)string_format(spam_score_buffer, sizeof(spam_score_buffer),
617 "%.1f", spamd_score);
618 spam_score = spam_score_buffer;
619
620 /* create "int" spam score */
621 j = (int)((spamd_score + 0.001)*10);
622 (void)string_format(spam_score_int_buffer, sizeof(spam_score_int_buffer),
623 "%d", j);
624 spam_score_int = spam_score_int_buffer;
625
626 /* compare threshold against score */
627 spam_rc = spamd_score >= spamd_threshold
628 ? OK /* spam as determined by user's threshold */
629 : FAIL; /* not spam */
630
631 /* remember expanded spamd_address if needed */
632 if (spamd_address_work != spamd_address)
633 prev_spamd_address_work = string_copy(spamd_address_work);
634
635 /* remember user name and "been here" for it */
636 Ustrcpy(prev_user_name, user_name);
637 spam_ok = 1;
638
639 return override
640 ? OK /* always return OK, no matter what the score */
641 : spam_rc;
642
643 defer:
644 (void)fclose(mbox_file);
645 return DEFER;
646 }
647
648 #endif
649 /* vi: aw ai sw=2
650 */